A new, officially sanctioned AI platform, EPGenAI Hub, is being rolled out by the European Parliament to standardize and govern AI use in legislative work, providing lawmakers and staff with access to models from Meta, OpenAI, Anthropic, and Mistral.

The Hub offers both locally hosted models (e.g., Meta’s Llama, OpenAI’s GPT-OSS, Mistral Small) and externally hosted options (including OpenAI’s ChatGPT and Anthropic’s Claude Sonnet), with demo access granted to MEPs and staff.

The initiative is compared with UK and US policies, with openness about vendors signaling a policy statement as much as a technology rollout.

The EP push for AI governance continues amid concerns about disinformation and reliability, seeking to protect democratic processes while enabling innovation.

Lawmakers Brando Benifei and Barry Andrews stress responsible AI use and the democratic implications of AI-assisted drafting.

The move seeks to balance efficient legislative work with safeguards for democratic processes, transparency, and data protection, with further changes anticipated.

An incident involving AI hallucination of a non-existent statute spurred the Hub, signaling a focus on process improvements rather than an outright ban on AI tools.

EU AI Act transparency requirements apply to interactions meant to mislead, but there is no current mandate to label legislators’ own AI-assisted work.

European users still favor American-language models, underscoring a tension between sovereignty goals and market dominance by US tech firms.

There is currently no requirement to disclose when AI is used to generate legislative questions or amendments, highlighting governance gaps even as the platform deploys.

Debate continues over governance and potential amendments to Parliament’s rules of procedure, with possible escalation to the Constitutional Affairs Committee for tighter regulation.

In line with digital sovereignty aims, the Parliament recently shifted its default search engine to the French Qwant to reduce US-dominated dependencies.